Solar Open 100B: benchmark results
Upstage's open-weight Solar Open (December 2025), a 102-billion-parameter, 12-billion-active MoE model trained from scratch on 19.7 trillion tokens for enterprise-grade reasoning and agentic tasks. Provider: Upstage. Released 2025-12-30. Access: Open.
Aggregate Rating 1610 ± 20, rank #406 of 1656 rated models, from 100 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| ALL Bench LLM | 75 | Average Numeric Benchmark Score (%) | 88.2 |
| Horangi 4 - Ko-MT-Bench | 88.86 | Judge rating (1-10, x10) | 84.6 |
| Upstage Solar Open 2 Card - HAE-RAE Bench v1.1 | 73.3 | Accuracy (%) | 80 |
| AA Omniscience - Software Engineering (SWE) - Dart | 24 | Accuracy (%) | 72.2 |
| AA Omniscience - Software Engineering (SWE) - Swift | 40 | Accuracy (%) | 70.1 |
| AA IFBench | 57.69 | Accuracy (%) | 69.3 |
| AA Omniscience - Software Engineering (SWE) - Kotlin | 22 | Accuracy (%) | 68.6 |
| AA Omniscience - Software Engineering (SWE) - Python | 23.5 | Accuracy (%) | 68.4 |
| AA Omniscience - Software Engineering (SWE) - R | 12 | Accuracy (%) | 62.1 |
| AA Omniscience - Software Engineering (SWE) - PHP | 24 | Accuracy (%) | 60.6 |
| Horangi 4 - IFEval-Ko | 86 | Instruction-following accuracy (%) | 59.6 |
| AA Omniscience - Software Engineering (SWE) - Java | 16 | Accuracy (%) | 57 |
Interactive version: theaggregate.ai/model?slug=solar-open-100b · How It Works · Data refreshed daily, snapshot 2026-10-20.